These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / gpu / drm / nouveau / nvkm / engine / gr / nv40.c
index 7e19379..05a8954 100644 (file)
 #include "regs.h"
 
 #include <core/client.h>
-#include <core/handle.h>
+#include <core/gpuobj.h>
 #include <subdev/fb.h>
 #include <subdev/timer.h>
 #include <engine/fifo.h>
 
-struct nv40_gr_priv {
-       struct nvkm_gr base;
-       u32 size;
-};
-
-struct nv40_gr_chan {
-       struct nvkm_gr_chan base;
-};
-
-static u64
+u64
 nv40_gr_units(struct nvkm_gr *gr)
 {
-       struct nv40_gr_priv *priv = (void *)gr;
-
-       return nv_rd32(priv, 0x1540);
+       return nvkm_rd32(gr->engine.subdev.device, 0x1540);
 }
 
 /*******************************************************************************
@@ -52,80 +41,29 @@ nv40_gr_units(struct nvkm_gr *gr)
  ******************************************************************************/
 
 static int
-nv40_gr_object_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-                   struct nvkm_oclass *oclass, void *data, u32 size,
-                   struct nvkm_object **pobject)
+nv40_gr_object_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
+                   int align, struct nvkm_gpuobj **pgpuobj)
 {
-       struct nvkm_gpuobj *obj;
-       int ret;
-
-       ret = nvkm_gpuobj_create(parent, engine, oclass, 0, parent,
-                                20, 16, 0, &obj);
-       *pobject = nv_object(obj);
-       if (ret)
-               return ret;
-
-       nv_wo32(obj, 0x00, nv_mclass(obj));
-       nv_wo32(obj, 0x04, 0x00000000);
-       nv_wo32(obj, 0x08, 0x00000000);
+       int ret = nvkm_gpuobj_new(object->engine->subdev.device, 20, align,
+                                 false, parent, pgpuobj);
+       if (ret == 0) {
+               nvkm_kmap(*pgpuobj);
+               nvkm_wo32(*pgpuobj, 0x00, object->oclass);
+               nvkm_wo32(*pgpuobj, 0x04, 0x00000000);
+               nvkm_wo32(*pgpuobj, 0x08, 0x00000000);
 #ifdef __BIG_ENDIAN
-       nv_mo32(obj, 0x08, 0x01000000, 0x01000000);
+               nvkm_mo32(*pgpuobj, 0x08, 0x01000000, 0x01000000);
 #endif
-       nv_wo32(obj, 0x0c, 0x00000000);
-       nv_wo32(obj, 0x10, 0x00000000);
-       return 0;
+               nvkm_wo32(*pgpuobj, 0x0c, 0x00000000);
+               nvkm_wo32(*pgpuobj, 0x10, 0x00000000);
+               nvkm_done(*pgpuobj);
+       }
+       return ret;
 }
 
-static struct nvkm_ofuncs
-nv40_gr_ofuncs = {
-       .ctor = nv40_gr_object_ctor,
-       .dtor = _nvkm_gpuobj_dtor,
-       .init = _nvkm_gpuobj_init,
-       .fini = _nvkm_gpuobj_fini,
-       .rd32 = _nvkm_gpuobj_rd32,
-       .wr32 = _nvkm_gpuobj_wr32,
-};
-
-static struct nvkm_oclass
-nv40_gr_sclass[] = {
-       { 0x0012, &nv40_gr_ofuncs, NULL }, /* beta1 */
-       { 0x0019, &nv40_gr_ofuncs, NULL }, /* clip */
-       { 0x0030, &nv40_gr_ofuncs, NULL }, /* null */
-       { 0x0039, &nv40_gr_ofuncs, NULL }, /* m2mf */
-       { 0x0043, &nv40_gr_ofuncs, NULL }, /* rop */
-       { 0x0044, &nv40_gr_ofuncs, NULL }, /* patt */
-       { 0x004a, &nv40_gr_ofuncs, NULL }, /* gdi */
-       { 0x0062, &nv40_gr_ofuncs, NULL }, /* surf2d */
-       { 0x0072, &nv40_gr_ofuncs, NULL }, /* beta4 */
-       { 0x0089, &nv40_gr_ofuncs, NULL }, /* sifm */
-       { 0x008a, &nv40_gr_ofuncs, NULL }, /* ifc */
-       { 0x009f, &nv40_gr_ofuncs, NULL }, /* imageblit */
-       { 0x3062, &nv40_gr_ofuncs, NULL }, /* surf2d (nv40) */
-       { 0x3089, &nv40_gr_ofuncs, NULL }, /* sifm (nv40) */
-       { 0x309e, &nv40_gr_ofuncs, NULL }, /* swzsurf (nv40) */
-       { 0x4097, &nv40_gr_ofuncs, NULL }, /* curie */
-       {},
-};
-
-static struct nvkm_oclass
-nv44_gr_sclass[] = {
-       { 0x0012, &nv40_gr_ofuncs, NULL }, /* beta1 */
-       { 0x0019, &nv40_gr_ofuncs, NULL }, /* clip */
-       { 0x0030, &nv40_gr_ofuncs, NULL }, /* null */
-       { 0x0039, &nv40_gr_ofuncs, NULL }, /* m2mf */
-       { 0x0043, &nv40_gr_ofuncs, NULL }, /* rop */
-       { 0x0044, &nv40_gr_ofuncs, NULL }, /* patt */
-       { 0x004a, &nv40_gr_ofuncs, NULL }, /* gdi */
-       { 0x0062, &nv40_gr_ofuncs, NULL }, /* surf2d */
-       { 0x0072, &nv40_gr_ofuncs, NULL }, /* beta4 */
-       { 0x0089, &nv40_gr_ofuncs, NULL }, /* sifm */
-       { 0x008a, &nv40_gr_ofuncs, NULL }, /* ifc */
-       { 0x009f, &nv40_gr_ofuncs, NULL }, /* imageblit */
-       { 0x3062, &nv40_gr_ofuncs, NULL }, /* surf2d (nv40) */
-       { 0x3089, &nv40_gr_ofuncs, NULL }, /* sifm (nv40) */
-       { 0x309e, &nv40_gr_ofuncs, NULL }, /* swzsurf (nv40) */
-       { 0x4497, &nv40_gr_ofuncs, NULL }, /* curie */
-       {},
+const struct nvkm_object_func
+nv40_gr_object = {
+       .bind = nv40_gr_object_bind,
 };
 
 /*******************************************************************************
@@ -133,361 +71,335 @@ nv44_gr_sclass[] = {
  ******************************************************************************/
 
 static int
-nv40_gr_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-                    struct nvkm_oclass *oclass, void *data, u32 size,
-                    struct nvkm_object **pobject)
+nv40_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent,
+                 int align, struct nvkm_gpuobj **pgpuobj)
 {
-       struct nv40_gr_priv *priv = (void *)engine;
-       struct nv40_gr_chan *chan;
-       int ret;
-
-       ret = nvkm_gr_context_create(parent, engine, oclass, NULL, priv->size,
-                                    16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
-       *pobject = nv_object(chan);
-       if (ret)
-               return ret;
-
-       nv40_grctx_fill(nv_device(priv), nv_gpuobj(chan));
-       nv_wo32(chan, 0x00000, nv_gpuobj(chan)->addr >> 4);
-       return 0;
+       struct nv40_gr_chan *chan = nv40_gr_chan(object);
+       struct nv40_gr *gr = chan->gr;
+       int ret = nvkm_gpuobj_new(gr->base.engine.subdev.device, gr->size,
+                                 align, true, parent, pgpuobj);
+       if (ret == 0) {
+               chan->inst = (*pgpuobj)->addr;
+               nvkm_kmap(*pgpuobj);
+               nv40_grctx_fill(gr->base.engine.subdev.device, *pgpuobj);
+               nvkm_wo32(*pgpuobj, 0x00000, chan->inst >> 4);
+               nvkm_done(*pgpuobj);
+       }
+       return ret;
 }
 
 static int
-nv40_gr_context_fini(struct nvkm_object *object, bool suspend)
+nv40_gr_chan_fini(struct nvkm_object *object, bool suspend)
 {
-       struct nv40_gr_priv *priv = (void *)object->engine;
-       struct nv40_gr_chan *chan = (void *)object;
-       u32 inst = 0x01000000 | nv_gpuobj(chan)->addr >> 4;
+       struct nv40_gr_chan *chan = nv40_gr_chan(object);
+       struct nv40_gr *gr = chan->gr;
+       struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+       struct nvkm_device *device = subdev->device;
+       u32 inst = 0x01000000 | chan->inst >> 4;
        int ret = 0;
 
-       nv_mask(priv, 0x400720, 0x00000001, 0x00000000);
+       nvkm_mask(device, 0x400720, 0x00000001, 0x00000000);
 
-       if (nv_rd32(priv, 0x40032c) == inst) {
+       if (nvkm_rd32(device, 0x40032c) == inst) {
                if (suspend) {
-                       nv_wr32(priv, 0x400720, 0x00000000);
-                       nv_wr32(priv, 0x400784, inst);
-                       nv_mask(priv, 0x400310, 0x00000020, 0x00000020);
-                       nv_mask(priv, 0x400304, 0x00000001, 0x00000001);
-                       if (!nv_wait(priv, 0x400300, 0x00000001, 0x00000000)) {
-                               u32 insn = nv_rd32(priv, 0x400308);
-                               nv_warn(priv, "ctxprog timeout 0x%08x\n", insn);
+                       nvkm_wr32(device, 0x400720, 0x00000000);
+                       nvkm_wr32(device, 0x400784, inst);
+                       nvkm_mask(device, 0x400310, 0x00000020, 0x00000020);
+                       nvkm_mask(device, 0x400304, 0x00000001, 0x00000001);
+                       if (nvkm_msec(device, 2000,
+                               if (!(nvkm_rd32(device, 0x400300) & 0x00000001))
+                                       break;
+                       ) < 0) {
+                               u32 insn = nvkm_rd32(device, 0x400308);
+                               nvkm_warn(subdev, "ctxprog timeout %08x\n", insn);
                                ret = -EBUSY;
                        }
                }
 
-               nv_mask(priv, 0x40032c, 0x01000000, 0x00000000);
+               nvkm_mask(device, 0x40032c, 0x01000000, 0x00000000);
        }
 
-       if (nv_rd32(priv, 0x400330) == inst)
-               nv_mask(priv, 0x400330, 0x01000000, 0x00000000);
+       if (nvkm_rd32(device, 0x400330) == inst)
+               nvkm_mask(device, 0x400330, 0x01000000, 0x00000000);
 
-       nv_mask(priv, 0x400720, 0x00000001, 0x00000001);
+       nvkm_mask(device, 0x400720, 0x00000001, 0x00000001);
        return ret;
 }
 
-static struct nvkm_oclass
-nv40_gr_cclass = {
-       .handle = NV_ENGCTX(GR, 0x40),
-       .ofuncs = &(struct nvkm_ofuncs) {
-               .ctor = nv40_gr_context_ctor,
-               .dtor = _nvkm_gr_context_dtor,
-               .init = _nvkm_gr_context_init,
-               .fini = nv40_gr_context_fini,
-               .rd32 = _nvkm_gr_context_rd32,
-               .wr32 = _nvkm_gr_context_wr32,
-       },
+static void *
+nv40_gr_chan_dtor(struct nvkm_object *object)
+{
+       struct nv40_gr_chan *chan = nv40_gr_chan(object);
+       unsigned long flags;
+       spin_lock_irqsave(&chan->gr->base.engine.lock, flags);
+       list_del(&chan->head);
+       spin_unlock_irqrestore(&chan->gr->base.engine.lock, flags);
+       return chan;
+}
+
+static const struct nvkm_object_func
+nv40_gr_chan = {
+       .dtor = nv40_gr_chan_dtor,
+       .fini = nv40_gr_chan_fini,
+       .bind = nv40_gr_chan_bind,
 };
 
+int
+nv40_gr_chan_new(struct nvkm_gr *base, struct nvkm_fifo_chan *fifoch,
+                const struct nvkm_oclass *oclass, struct nvkm_object **pobject)
+{
+       struct nv40_gr *gr = nv40_gr(base);
+       struct nv40_gr_chan *chan;
+       unsigned long flags;
+
+       if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL)))
+               return -ENOMEM;
+       nvkm_object_ctor(&nv40_gr_chan, oclass, &chan->object);
+       chan->gr = gr;
+       chan->fifo = fifoch;
+       *pobject = &chan->object;
+
+       spin_lock_irqsave(&chan->gr->base.engine.lock, flags);
+       list_add(&chan->head, &gr->chan);
+       spin_unlock_irqrestore(&chan->gr->base.engine.lock, flags);
+       return 0;
+}
+
 /*******************************************************************************
  * PGRAPH engine/subdev functions
  ******************************************************************************/
 
 static void
-nv40_gr_tile_prog(struct nvkm_engine *engine, int i)
+nv40_gr_tile(struct nvkm_gr *base, int i, struct nvkm_fb_tile *tile)
 {
-       struct nvkm_fb_tile *tile = &nvkm_fb(engine)->tile.region[i];
-       struct nvkm_fifo *pfifo = nvkm_fifo(engine);
-       struct nv40_gr_priv *priv = (void *)engine;
+       struct nv40_gr *gr = nv40_gr(base);
+       struct nvkm_device *device = gr->base.engine.subdev.device;
+       struct nvkm_fifo *fifo = device->fifo;
        unsigned long flags;
 
-       pfifo->pause(pfifo, &flags);
-       nv04_gr_idle(priv);
+       nvkm_fifo_pause(fifo, &flags);
+       nv04_gr_idle(&gr->base);
 
-       switch (nv_device(priv)->chipset) {
+       switch (device->chipset) {
        case 0x40:
        case 0x41:
        case 0x42:
        case 0x43:
        case 0x45:
-       case 0x4e:
-               nv_wr32(priv, NV20_PGRAPH_TSIZE(i), tile->pitch);
-               nv_wr32(priv, NV20_PGRAPH_TLIMIT(i), tile->limit);
-               nv_wr32(priv, NV20_PGRAPH_TILE(i), tile->addr);
-               nv_wr32(priv, NV40_PGRAPH_TSIZE1(i), tile->pitch);
-               nv_wr32(priv, NV40_PGRAPH_TLIMIT1(i), tile->limit);
-               nv_wr32(priv, NV40_PGRAPH_TILE1(i), tile->addr);
-               switch (nv_device(priv)->chipset) {
+               nvkm_wr32(device, NV20_PGRAPH_TSIZE(i), tile->pitch);
+               nvkm_wr32(device, NV20_PGRAPH_TLIMIT(i), tile->limit);
+               nvkm_wr32(device, NV20_PGRAPH_TILE(i), tile->addr);
+               nvkm_wr32(device, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+               nvkm_wr32(device, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+               nvkm_wr32(device, NV40_PGRAPH_TILE1(i), tile->addr);
+               switch (device->chipset) {
                case 0x40:
                case 0x45:
-                       nv_wr32(priv, NV20_PGRAPH_ZCOMP(i), tile->zcomp);
-                       nv_wr32(priv, NV40_PGRAPH_ZCOMP1(i), tile->zcomp);
+                       nvkm_wr32(device, NV20_PGRAPH_ZCOMP(i), tile->zcomp);
+                       nvkm_wr32(device, NV40_PGRAPH_ZCOMP1(i), tile->zcomp);
                        break;
                case 0x41:
                case 0x42:
                case 0x43:
-                       nv_wr32(priv, NV41_PGRAPH_ZCOMP0(i), tile->zcomp);
-                       nv_wr32(priv, NV41_PGRAPH_ZCOMP1(i), tile->zcomp);
+                       nvkm_wr32(device, NV41_PGRAPH_ZCOMP0(i), tile->zcomp);
+                       nvkm_wr32(device, NV41_PGRAPH_ZCOMP1(i), tile->zcomp);
                        break;
                default:
                        break;
                }
                break;
-       case 0x44:
-       case 0x4a:
-               nv_wr32(priv, NV20_PGRAPH_TSIZE(i), tile->pitch);
-               nv_wr32(priv, NV20_PGRAPH_TLIMIT(i), tile->limit);
-               nv_wr32(priv, NV20_PGRAPH_TILE(i), tile->addr);
-               break;
-       case 0x46:
-       case 0x4c:
        case 0x47:
        case 0x49:
        case 0x4b:
-       case 0x63:
-       case 0x67:
-       case 0x68:
-               nv_wr32(priv, NV47_PGRAPH_TSIZE(i), tile->pitch);
-               nv_wr32(priv, NV47_PGRAPH_TLIMIT(i), tile->limit);
-               nv_wr32(priv, NV47_PGRAPH_TILE(i), tile->addr);
-               nv_wr32(priv, NV40_PGRAPH_TSIZE1(i), tile->pitch);
-               nv_wr32(priv, NV40_PGRAPH_TLIMIT1(i), tile->limit);
-               nv_wr32(priv, NV40_PGRAPH_TILE1(i), tile->addr);
-               switch (nv_device(priv)->chipset) {
-               case 0x47:
-               case 0x49:
-               case 0x4b:
-                       nv_wr32(priv, NV47_PGRAPH_ZCOMP0(i), tile->zcomp);
-                       nv_wr32(priv, NV47_PGRAPH_ZCOMP1(i), tile->zcomp);
-                       break;
-               default:
-                       break;
-               }
+               nvkm_wr32(device, NV47_PGRAPH_TSIZE(i), tile->pitch);
+               nvkm_wr32(device, NV47_PGRAPH_TLIMIT(i), tile->limit);
+               nvkm_wr32(device, NV47_PGRAPH_TILE(i), tile->addr);
+               nvkm_wr32(device, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+               nvkm_wr32(device, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+               nvkm_wr32(device, NV40_PGRAPH_TILE1(i), tile->addr);
+               nvkm_wr32(device, NV47_PGRAPH_ZCOMP0(i), tile->zcomp);
+               nvkm_wr32(device, NV47_PGRAPH_ZCOMP1(i), tile->zcomp);
                break;
        default:
+               WARN_ON(1);
                break;
        }
 
-       pfifo->start(pfifo, &flags);
+       nvkm_fifo_start(fifo, &flags);
 }
 
-static void
-nv40_gr_intr(struct nvkm_subdev *subdev)
+void
+nv40_gr_intr(struct nvkm_gr *base)
 {
-       struct nvkm_fifo *pfifo = nvkm_fifo(subdev);
-       struct nvkm_engine *engine = nv_engine(subdev);
-       struct nvkm_object *engctx;
-       struct nvkm_handle *handle = NULL;
-       struct nv40_gr_priv *priv = (void *)subdev;
-       u32 stat = nv_rd32(priv, NV03_PGRAPH_INTR);
-       u32 nsource = nv_rd32(priv, NV03_PGRAPH_NSOURCE);
-       u32 nstatus = nv_rd32(priv, NV03_PGRAPH_NSTATUS);
-       u32 inst = nv_rd32(priv, 0x40032c) & 0x000fffff;
-       u32 addr = nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR);
+       struct nv40_gr *gr = nv40_gr(base);
+       struct nv40_gr_chan *temp, *chan = NULL;
+       struct nvkm_subdev *subdev = &gr->base.engine.subdev;
+       struct nvkm_device *device = subdev->device;
+       u32 stat = nvkm_rd32(device, NV03_PGRAPH_INTR);
+       u32 nsource = nvkm_rd32(device, NV03_PGRAPH_NSOURCE);
+       u32 nstatus = nvkm_rd32(device, NV03_PGRAPH_NSTATUS);
+       u32 inst = nvkm_rd32(device, 0x40032c) & 0x000fffff;
+       u32 addr = nvkm_rd32(device, NV04_PGRAPH_TRAPPED_ADDR);
        u32 subc = (addr & 0x00070000) >> 16;
        u32 mthd = (addr & 0x00001ffc);
-       u32 data = nv_rd32(priv, NV04_PGRAPH_TRAPPED_DATA);
-       u32 class = nv_rd32(priv, 0x400160 + subc * 4) & 0xffff;
+       u32 data = nvkm_rd32(device, NV04_PGRAPH_TRAPPED_DATA);
+       u32 class = nvkm_rd32(device, 0x400160 + subc * 4) & 0xffff;
        u32 show = stat;
-       int chid;
-
-       engctx = nvkm_engctx_get(engine, inst);
-       chid   = pfifo->chid(pfifo, engctx);
+       char msg[128], src[128], sta[128];
+       unsigned long flags;
 
-       if (stat & NV_PGRAPH_INTR_ERROR) {
-               if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
-                       handle = nvkm_handle_get_class(engctx, class);
-                       if (handle && !nv_call(handle->object, mthd, data))
-                               show &= ~NV_PGRAPH_INTR_ERROR;
-                       nvkm_handle_put(handle);
+       spin_lock_irqsave(&gr->base.engine.lock, flags);
+       list_for_each_entry(temp, &gr->chan, head) {
+               if (temp->inst >> 4 == inst) {
+                       chan = temp;
+                       list_del(&chan->head);
+                       list_add(&chan->head, &gr->chan);
+                       break;
                }
+       }
 
+       if (stat & NV_PGRAPH_INTR_ERROR) {
                if (nsource & NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION) {
-                       nv_mask(priv, 0x402000, 0, 0);
+                       nvkm_mask(device, 0x402000, 0, 0);
                }
        }
 
-       nv_wr32(priv, NV03_PGRAPH_INTR, stat);
-       nv_wr32(priv, NV04_PGRAPH_FIFO, 0x00000001);
+       nvkm_wr32(device, NV03_PGRAPH_INTR, stat);
+       nvkm_wr32(device, NV04_PGRAPH_FIFO, 0x00000001);
 
        if (show) {
-               nv_error(priv, "%s", "");
-               nvkm_bitfield_print(nv10_gr_intr_name, show);
-               pr_cont(" nsource:");
-               nvkm_bitfield_print(nv04_gr_nsource, nsource);
-               pr_cont(" nstatus:");
-               nvkm_bitfield_print(nv10_gr_nstatus, nstatus);
-               pr_cont("\n");
-               nv_error(priv,
-                        "ch %d [0x%08x %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x\n",
-                        chid, inst << 4, nvkm_client_name(engctx), subc,
-                        class, mthd, data);
+               nvkm_snprintbf(msg, sizeof(msg), nv10_gr_intr_name, show);
+               nvkm_snprintbf(src, sizeof(src), nv04_gr_nsource, nsource);
+               nvkm_snprintbf(sta, sizeof(sta), nv10_gr_nstatus, nstatus);
+               nvkm_error(subdev, "intr %08x [%s] nsource %08x [%s] "
+                                  "nstatus %08x [%s] ch %d [%08x %s] subc %d "
+                                  "class %04x mthd %04x data %08x\n",
+                          show, msg, nsource, src, nstatus, sta,
+                          chan ? chan->fifo->chid : -1, inst << 4,
+                          chan ? chan->fifo->object.client->name : "unknown",
+                          subc, class, mthd, data);
        }
 
-       nvkm_engctx_put(engctx);
+       spin_unlock_irqrestore(&gr->base.engine.lock, flags);
 }
 
-static int
-nv40_gr_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
-            struct nvkm_oclass *oclass, void *data, u32 size,
-            struct nvkm_object **pobject)
+int
+nv40_gr_init(struct nvkm_gr *base)
 {
-       struct nv40_gr_priv *priv;
-       int ret;
-
-       ret = nvkm_gr_create(parent, engine, oclass, true, &priv);
-       *pobject = nv_object(priv);
-       if (ret)
-               return ret;
-
-       nv_subdev(priv)->unit = 0x00001000;
-       nv_subdev(priv)->intr = nv40_gr_intr;
-       nv_engine(priv)->cclass = &nv40_gr_cclass;
-       if (nv44_gr_class(priv))
-               nv_engine(priv)->sclass = nv44_gr_sclass;
-       else
-               nv_engine(priv)->sclass = nv40_gr_sclass;
-       nv_engine(priv)->tile_prog = nv40_gr_tile_prog;
-
-       priv->base.units = nv40_gr_units;
-       return 0;
-}
-
-static int
-nv40_gr_init(struct nvkm_object *object)
-{
-       struct nvkm_engine *engine = nv_engine(object);
-       struct nvkm_fb *pfb = nvkm_fb(object);
-       struct nv40_gr_priv *priv = (void *)engine;
+       struct nv40_gr *gr = nv40_gr(base);
+       struct nvkm_device *device = gr->base.engine.subdev.device;
        int ret, i, j;
        u32 vramsz;
 
-       ret = nvkm_gr_init(&priv->base);
-       if (ret)
-               return ret;
-
        /* generate and upload context program */
-       ret = nv40_grctx_init(nv_device(priv), &priv->size);
+       ret = nv40_grctx_init(device, &gr->size);
        if (ret)
                return ret;
 
        /* No context present currently */
-       nv_wr32(priv, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
+       nvkm_wr32(device, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
 
-       nv_wr32(priv, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
-       nv_wr32(priv, NV40_PGRAPH_INTR_EN, 0xFFFFFFFF);
+       nvkm_wr32(device, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
+       nvkm_wr32(device, NV40_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
-       nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
-       nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x00000000);
-       nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x401287c0);
-       nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xe0de8055);
-       nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x00008000);
-       nv_wr32(priv, NV04_PGRAPH_LIMIT_VIOL_PIX, 0x00be3c5f);
+       nvkm_wr32(device, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
+       nvkm_wr32(device, NV04_PGRAPH_DEBUG_0, 0x00000000);
+       nvkm_wr32(device, NV04_PGRAPH_DEBUG_1, 0x401287c0);
+       nvkm_wr32(device, NV04_PGRAPH_DEBUG_3, 0xe0de8055);
+       nvkm_wr32(device, NV10_PGRAPH_DEBUG_4, 0x00008000);
+       nvkm_wr32(device, NV04_PGRAPH_LIMIT_VIOL_PIX, 0x00be3c5f);
 
-       nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10010100);
-       nv_wr32(priv, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
+       nvkm_wr32(device, NV10_PGRAPH_CTX_CONTROL, 0x10010100);
+       nvkm_wr32(device, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
 
-       j = nv_rd32(priv, 0x1540) & 0xff;
+       j = nvkm_rd32(device, 0x1540) & 0xff;
        if (j) {
                for (i = 0; !(j & 1); j >>= 1, i++)
                        ;
-               nv_wr32(priv, 0x405000, i);
+               nvkm_wr32(device, 0x405000, i);
        }
 
-       if (nv_device(priv)->chipset == 0x40) {
-               nv_wr32(priv, 0x4009b0, 0x83280fff);
-               nv_wr32(priv, 0x4009b4, 0x000000a0);
+       if (device->chipset == 0x40) {
+               nvkm_wr32(device, 0x4009b0, 0x83280fff);
+               nvkm_wr32(device, 0x4009b4, 0x000000a0);
        } else {
-               nv_wr32(priv, 0x400820, 0x83280eff);
-               nv_wr32(priv, 0x400824, 0x000000a0);
+               nvkm_wr32(device, 0x400820, 0x83280eff);
+               nvkm_wr32(device, 0x400824, 0x000000a0);
        }
 
-       switch (nv_device(priv)->chipset) {
+       switch (device->chipset) {
        case 0x40:
        case 0x45:
-               nv_wr32(priv, 0x4009b8, 0x0078e366);
-               nv_wr32(priv, 0x4009bc, 0x0000014c);
+               nvkm_wr32(device, 0x4009b8, 0x0078e366);
+               nvkm_wr32(device, 0x4009bc, 0x0000014c);
                break;
        case 0x41:
        case 0x42: /* pciid also 0x00Cx */
        /* case 0x0120: XXX (pciid) */
-               nv_wr32(priv, 0x400828, 0x007596ff);
-               nv_wr32(priv, 0x40082c, 0x00000108);
+               nvkm_wr32(device, 0x400828, 0x007596ff);
+               nvkm_wr32(device, 0x40082c, 0x00000108);
                break;
        case 0x43:
-               nv_wr32(priv, 0x400828, 0x0072cb77);
-               nv_wr32(priv, 0x40082c, 0x00000108);
+               nvkm_wr32(device, 0x400828, 0x0072cb77);
+               nvkm_wr32(device, 0x40082c, 0x00000108);
                break;
        case 0x44:
        case 0x46: /* G72 */
        case 0x4a:
        case 0x4c: /* G7x-based C51 */
        case 0x4e:
-               nv_wr32(priv, 0x400860, 0);
-               nv_wr32(priv, 0x400864, 0);
+               nvkm_wr32(device, 0x400860, 0);
+               nvkm_wr32(device, 0x400864, 0);
                break;
        case 0x47: /* G70 */
        case 0x49: /* G71 */
        case 0x4b: /* G73 */
-               nv_wr32(priv, 0x400828, 0x07830610);
-               nv_wr32(priv, 0x40082c, 0x0000016A);
+               nvkm_wr32(device, 0x400828, 0x07830610);
+               nvkm_wr32(device, 0x40082c, 0x0000016A);
                break;
        default:
                break;
        }
 
-       nv_wr32(priv, 0x400b38, 0x2ffff800);
-       nv_wr32(priv, 0x400b3c, 0x00006000);
+       nvkm_wr32(device, 0x400b38, 0x2ffff800);
+       nvkm_wr32(device, 0x400b3c, 0x00006000);
 
        /* Tiling related stuff. */
-       switch (nv_device(priv)->chipset) {
+       switch (device->chipset) {
        case 0x44:
        case 0x4a:
-               nv_wr32(priv, 0x400bc4, 0x1003d888);
-               nv_wr32(priv, 0x400bbc, 0xb7a7b500);
+               nvkm_wr32(device, 0x400bc4, 0x1003d888);
+               nvkm_wr32(device, 0x400bbc, 0xb7a7b500);
                break;
        case 0x46:
-               nv_wr32(priv, 0x400bc4, 0x0000e024);
-               nv_wr32(priv, 0x400bbc, 0xb7a7b520);
+               nvkm_wr32(device, 0x400bc4, 0x0000e024);
+               nvkm_wr32(device, 0x400bbc, 0xb7a7b520);
                break;
        case 0x4c:
        case 0x4e:
        case 0x67:
-               nv_wr32(priv, 0x400bc4, 0x1003d888);
-               nv_wr32(priv, 0x400bbc, 0xb7a7b540);
+               nvkm_wr32(device, 0x400bc4, 0x1003d888);
+               nvkm_wr32(device, 0x400bbc, 0xb7a7b540);
                break;
        default:
                break;
        }
 
-       /* Turn all the tiling regions off. */
-       for (i = 0; i < pfb->tile.regions; i++)
-               engine->tile_prog(engine, i);
-
        /* begin RAM config */
-       vramsz = nv_device_resource_len(nv_device(priv), 0) - 1;
-       switch (nv_device(priv)->chipset) {
+       vramsz = device->func->resource_size(device, 1) - 1;
+       switch (device->chipset) {
        case 0x40:
-               nv_wr32(priv, 0x4009A4, nv_rd32(priv, 0x100200));
-               nv_wr32(priv, 0x4009A8, nv_rd32(priv, 0x100204));
-               nv_wr32(priv, 0x4069A4, nv_rd32(priv, 0x100200));
-               nv_wr32(priv, 0x4069A8, nv_rd32(priv, 0x100204));
-               nv_wr32(priv, 0x400820, 0);
-               nv_wr32(priv, 0x400824, 0);
-               nv_wr32(priv, 0x400864, vramsz);
-               nv_wr32(priv, 0x400868, vramsz);
+               nvkm_wr32(device, 0x4009A4, nvkm_rd32(device, 0x100200));
+               nvkm_wr32(device, 0x4009A8, nvkm_rd32(device, 0x100204));
+               nvkm_wr32(device, 0x4069A4, nvkm_rd32(device, 0x100200));
+               nvkm_wr32(device, 0x4069A8, nvkm_rd32(device, 0x100204));
+               nvkm_wr32(device, 0x400820, 0);
+               nvkm_wr32(device, 0x400824, 0);
+               nvkm_wr32(device, 0x400864, vramsz);
+               nvkm_wr32(device, 0x400868, vramsz);
                break;
        default:
-               switch (nv_device(priv)->chipset) {
+               switch (device->chipset) {
                case 0x41:
                case 0x42:
                case 0x43:
@@ -495,33 +407,70 @@ nv40_gr_init(struct nvkm_object *object)
                case 0x4e:
                case 0x44:
                case 0x4a:
-                       nv_wr32(priv, 0x4009F0, nv_rd32(priv, 0x100200));
-                       nv_wr32(priv, 0x4009F4, nv_rd32(priv, 0x100204));
+                       nvkm_wr32(device, 0x4009F0, nvkm_rd32(device, 0x100200));
+                       nvkm_wr32(device, 0x4009F4, nvkm_rd32(device, 0x100204));
                        break;
                default:
-                       nv_wr32(priv, 0x400DF0, nv_rd32(priv, 0x100200));
-                       nv_wr32(priv, 0x400DF4, nv_rd32(priv, 0x100204));
+                       nvkm_wr32(device, 0x400DF0, nvkm_rd32(device, 0x100200));
+                       nvkm_wr32(device, 0x400DF4, nvkm_rd32(device, 0x100204));
                        break;
                }
-               nv_wr32(priv, 0x4069F0, nv_rd32(priv, 0x100200));
-               nv_wr32(priv, 0x4069F4, nv_rd32(priv, 0x100204));
-               nv_wr32(priv, 0x400840, 0);
-               nv_wr32(priv, 0x400844, 0);
-               nv_wr32(priv, 0x4008A0, vramsz);
-               nv_wr32(priv, 0x4008A4, vramsz);
+               nvkm_wr32(device, 0x4069F0, nvkm_rd32(device, 0x100200));
+               nvkm_wr32(device, 0x4069F4, nvkm_rd32(device, 0x100204));
+               nvkm_wr32(device, 0x400840, 0);
+               nvkm_wr32(device, 0x400844, 0);
+               nvkm_wr32(device, 0x4008A0, vramsz);
+               nvkm_wr32(device, 0x4008A4, vramsz);
                break;
        }
 
        return 0;
 }
 
-struct nvkm_oclass
-nv40_gr_oclass = {
-       .handle = NV_ENGINE(GR, 0x40),
-       .ofuncs = &(struct nvkm_ofuncs) {
-               .ctor = nv40_gr_ctor,
-               .dtor = _nvkm_gr_dtor,
-               .init = nv40_gr_init,
-               .fini = _nvkm_gr_fini,
-       },
+int
+nv40_gr_new_(const struct nvkm_gr_func *func, struct nvkm_device *device,
+            int index, struct nvkm_gr **pgr)
+{
+       struct nv40_gr *gr;
+
+       if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL)))
+               return -ENOMEM;
+       *pgr = &gr->base;
+       INIT_LIST_HEAD(&gr->chan);
+
+       return nvkm_gr_ctor(func, device, index, 0x00001000, true, &gr->base);
+}
+
+static const struct nvkm_gr_func
+nv40_gr = {
+       .init = nv40_gr_init,
+       .intr = nv40_gr_intr,
+       .tile = nv40_gr_tile,
+       .units = nv40_gr_units,
+       .chan_new = nv40_gr_chan_new,
+       .sclass = {
+               { -1, -1, 0x0012, &nv40_gr_object }, /* beta1 */
+               { -1, -1, 0x0019, &nv40_gr_object }, /* clip */
+               { -1, -1, 0x0030, &nv40_gr_object }, /* null */
+               { -1, -1, 0x0039, &nv40_gr_object }, /* m2mf */
+               { -1, -1, 0x0043, &nv40_gr_object }, /* rop */
+               { -1, -1, 0x0044, &nv40_gr_object }, /* patt */
+               { -1, -1, 0x004a, &nv40_gr_object }, /* gdi */
+               { -1, -1, 0x0062, &nv40_gr_object }, /* surf2d */
+               { -1, -1, 0x0072, &nv40_gr_object }, /* beta4 */
+               { -1, -1, 0x0089, &nv40_gr_object }, /* sifm */
+               { -1, -1, 0x008a, &nv40_gr_object }, /* ifc */
+               { -1, -1, 0x009f, &nv40_gr_object }, /* imageblit */
+               { -1, -1, 0x3062, &nv40_gr_object }, /* surf2d (nv40) */
+               { -1, -1, 0x3089, &nv40_gr_object }, /* sifm (nv40) */
+               { -1, -1, 0x309e, &nv40_gr_object }, /* swzsurf (nv40) */
+               { -1, -1, 0x4097, &nv40_gr_object }, /* curie */
+               {}
+       }
 };
+
+int
+nv40_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
+{
+       return nv40_gr_new_(&nv40_gr, device, index, pgr);
+}